Whispering now supports completely offline transcription using Whisper.cpp, introduced in #655. No API keys, no internet connection required. Just download a model and transcribe directly on your device with the same user experience as cloud services.
Select "Whisper C++" from the transcription dropdown, download a pre-configured model (such as Tiny at 78MB or Small at 488MB), and start transcribing locally. We recommend installing FFmpeg, as Whispering uses this under the hood to handle audio format conversion when needed (Whisper C++ models only support 16 kHz mono audio files).
For Existing Users
If you're upgrading from a previous version, we recommend changing two new features:
Switch to Native Recording Backend: Head to Settings → Recording and try switching from "Browser" to "Native" backend. The native backend provides better performance and reliability, especially on macOS and Windows.
Try Local Transcription: Download the Whisper C++ Small model (488MB) from Settings → Transcription and test local transcription. It works completely offline and keeps your audio private while providing excellent accuracy for most use cases.
What's Changed
- Convert prettier config files to JS for auto-format from VS Code. by @Leftium in #662
- feat: transcription selector popover ui fix by @thisisharsh7 in #660
- feat: add Deepgram nova-3 model by @heliole in #666
- feat: add macOS permissions integration by @braden-w in #668
- refactor: rename clipboard service to text service with independent operations by @braden-w in #659
- feat: add Whisper.cpp support for truly local transcription by @braden-w in #655
New Contributors
Full Changelog: v7.2.2...v7.3.0
Questions? Join our Discord or check the documentation.
Love Whispering? Star us on GitHub to show your support!